Open Source
Lexaloud
Neural text-to-speech daemon for academic reading-along on Linux
About Lexaloud
HAVEN Intelligence developed Lexaloud, an open source text-to-speech daemon designed for academic reading-along on Linux. The tool demonstrates our approach to privacy-first AI solutions: a neural voice model runs locally on GPU, no data leaves the machine, no telemetry is collected.
Lexaloud uses the Kokoro-82M voice model via ONNX Runtime to generate natural speech in real time. The user selects text, presses a global hotkey, and Lexaloud reads aloud with sentence-granularity and full control over pause, skip, and back.
Architecture
Lexaloud is built as a Unix domain socket daemon with a GTK3 tray indicator and control window. The architecture separates inference from the user interface, so the voice model runs stably in the background.
Component 1
Daemon (Unix socket)
A background service listening on a Unix domain socket. The daemon receives text, splits it into sentences, and streams speech via Kokoro-82M. Can be managed with systemd or started manually.
Component 2
Kokoro-82M via ONNX Runtime
A neural voice model with 82 million parameters runs locally on GPU. ONNX Runtime ensures fast inference with low latency, generating speech in real time with natural prosody.
Component 3
GTK3 tray indicator and control window
A system tray indicator provides a quick status overview. The control window shows the current sentence and gives access to pause, skip, and back. A global hotkey activates playback from any application.
Key Features
- ■ Kokoro-82M neural voice model on local GPU via ONNX Runtime
- ■ Sentence-granularity streaming with pause, skip, and back
- ■ Global hotkey that works from any application
- ■ GTK3 tray indicator and control window
- ■ Multi-distro support: Ubuntu, Fedora, and Arch Linux
- ■ Privacy: no telemetry, no cloud, no data collection
- ■ Open source under MIT license
Source Code
Lexaloud is open source and available on GitHub. Browse the code, read the documentation, or contribute to the project.
View on GitHubNeed practical AI solutions?
Use the case as technical reference, but start new projects with an advisory clarification of workflows, data, and risk.
Book a 20-min discovery callPersonal reply from Gustav. No automated sales sequence.